fix (#4475)

Co-authored-by: tastelikefeet <yuze.zyz@alibaab-inc.com>

Author: tastelikefeet

examples/train/multi-gpu/fsdp_qlora/train.sh

@@ -4,7 +4,7 @@ nproc_per_node=2
 CUDA_VISIBLE_DEVICES=0,1 \
 accelerate launch --config_file "./examples/train/multi-gpu/fsdp_qlora/fsdp_offload.json" \
     swift/cli/sft.py \
-    --model Qwen/Qwen2.5-7B-Instruct \
+    --model Qwen/Qwen2.5-72B-Instruct \
     --train_type lora \
     --dataset 'swift/self-cognition#1000' \
     --torch_dtype bfloat16 \

swift/plugin/prm.py

@@ -54,14 +54,14 @@ def __call__(self, infer_requests: List[Union[InferRequest, Dict]], ground_truth
         )
 
         for request, ground_truth in zip(infer_requests, ground_truths):
-            previous = request['messages'][:-1]
+            previous = request.messages[:-1]
             if previous[0]['role'] == 'system':
                 previous = previous[1:]
 
-            assert request['messages'][-1]['role'] == 'assistant'
+            assert request.messages[-1]['role'] == 'assistant'
             query = QUERY.replace('#query#', json.dumps(previous))
             query = query.replace('#ground_truth#', ground_truth)
-            query = query.replace('#response#', request['messages'][-1]['content'])
+            query = query.replace('#response#', request.messages[-1]['content'])
             messages = [
                 {
                     'role': 'system',